Beamforming may be used to improve signal-to-noise ratio, reduce reverberation, cancel interference, or estimate source location. It is done with an array of sensors using a model of the arriving wavefront. Conventionally, a plane wave or point source model is used and this can cause decreased array gain or even total breakdown of beamforming when the source is directional. The directivity of some real world acoustic sources is reviewed in this work. Examples are shown where using a point source model leads to large performance penalties, such as when the array subtends a significant angle about the source. The performance penalty of using a point source model to beamform on the human talker is estimated using phase information available in the literature.;Methods are demonstrated to estimate the weighted sum, based on the observed data. Simulation and experimentally measured results for a simple directional source and an array of 16 microphones show that the proposed beamforming method provides an array gain of about 11 dB while conventional beamforming, with a value of -4 dB, does not provide any useful gain. Simulation results for simple sources in the presence of diffuse noise, diffuse noise and reverberation, or diffuse noise and an interferer are also presented. The computational complexity of the proposed beamforming method is also compared to that of conventional beamforming.;An extension of beamforming applicable to directional sources with unknown directivity is presented. In the thesis, the proposed method is mostly intended for sound pickup of audio sources. The proposed beamforming method defines a set of "sub-beamformers", each designed to respond to a different spatial mode of the source. The outputs of the individual sub-beamformers are combined in a weighted sum to give an overall output of better quality than that of a conventional (point source) beamformer. It is shown that with appropriate weighting, the optimum array gain can be achieved In the case that the source happens to be a point source, the method reduces to conventional beamforming.
展开▼
